[PATCH v2 0/4] replace system_wq with system_dfl_wq, add WQ_UNBOUND to alloc_workqueue

[PATCH v2 0/4] replace system_wq with system_dfl_wq, add WQ_UNBOUND to alloc_workqueue

[Marco Crivellari]

Hi,

=== Current situation: problems ===

Let's consider a nohz_full system with isolated CPUs: wq_unbound_cpumask is
set to the housekeeping CPUs, for !WQ_UNBOUND the local CPU is selected.

This leads to different scenarios if a work item is scheduled on an
isolated CPU where "delay" value is 0 or greater then 0:
        schedule_delayed_work(, 0);

This will be handled by __queue_work() that will queue the work item on the
current local (isolated) CPU, while:

        schedule_delayed_work(, 1);

Will move the timer on an housekeeping CPU, and schedule the work there.

Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.

This lack of consistency cannot be addressed without refactoring the API.

=== Recent changes to the WQ API ===

The following, address the recent changes in the Workqueue API:

- commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
- commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")

The old workqueues will be removed in a future release cycle.

=== Introduced Changes by this series ===

1) [P 1-2-3]  Replace uses of system_wq

    system_wq is a per-CPU workqueue, these workloads do not benefit
	from locality, so change the the workqueue with the unbound,
	system_dfl_wq.

2) [P 4] add WQ_UNBOUND to alloc_workqueue()

	alloc_workqueue() users must specify one among WQ_PERCPU or WQ_UNBOUND.
	This workload does not benefit from a per-cpu workqueue, so make the
	custom wq unbound with WQ_UNBOUND.

Thanks!

---
Changes in v2:
- all the patches use system_dfl_wq and the custom wq is now unbound,
  using explicitly WQ_UNBOUND.

Marco Crivellari (4):
  Input: gpio_keys - replace use of system_wq with system_dfl_wq
  Input: palmas-pwrbutton - replace use of system_wq with system_dfl_wq
  Input: mouse - replace use of system_wq with system_dfl_wq
  Input: psmouse-smbus - add WQ_UNBOUND to alloc_workqueue user

 drivers/input/keyboard/gpio_keys.c    | 2 +-
 drivers/input/misc/palmas-pwrbutton.c | 2 +-
 drivers/input/mouse/psmouse-smbus.c   | 2 +-
 drivers/input/mouse/synaptics_i2c.c   | 8 ++++----
 4 files changed, 7 insertions(+), 7 deletions(-)

-- 
2.51.1

[PATCH v2 1/4] Input: gpio_keys - replace use of system_wq with system_dfl_wq

[Marco Crivellari]

Currently if a user enqueues a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.

This lack of consistency cannot be addressed without refactoring the API.

This patch continues the effort to refactor worqueue APIs, which has begun
with the change introducing new workqueues and a new alloc_workqueue flag:

commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")

This specific workload do not benefit from a per-cpu workqueue, so use
the default unbound workqueue (system_dfl_wq) instead.

Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
---
 drivers/input/keyboard/gpio_keys.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/input/keyboard/gpio_keys.c b/drivers/input/keyboard/gpio_keys.c
index f9db86da0818..5e875771a066 100644
--- a/drivers/input/keyboard/gpio_keys.c
+++ b/drivers/input/keyboard/gpio_keys.c
@@ -434,7 +434,7 @@ static irqreturn_t gpio_keys_gpio_isr(int irq, void *dev_id)
 			      ms_to_ktime(bdata->software_debounce),
 			      HRTIMER_MODE_REL);
 	} else {
-		mod_delayed_work(system_wq,
+		mod_delayed_work(system_dfl_wq,
 				 &bdata->work,
 				 msecs_to_jiffies(bdata->software_debounce));
 	}
-- 
2.51.1

[PATCH v2 2/4] Input: palmas-pwrbutton - replace use of system_wq with system_dfl_wq

[Marco Crivellari]

Currently if a user enqueues a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.

This lack of consistency cannot be addressed without refactoring the API.

This patch continues the effort to refactor worqueue APIs, which has begun
with the change introducing new workqueues and a new alloc_workqueue flag:

commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")

This specific workload do not benefit from a per-cpu workqueue, so use
the default unbound workqueue (system_dfl_wq) instead.

Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
---
 drivers/input/misc/palmas-pwrbutton.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/input/misc/palmas-pwrbutton.c b/drivers/input/misc/palmas-pwrbutton.c
index 39fc451c56e9..d9c5aae143dc 100644
--- a/drivers/input/misc/palmas-pwrbutton.c
+++ b/drivers/input/misc/palmas-pwrbutton.c
@@ -91,7 +91,7 @@ static irqreturn_t pwron_irq(int irq, void *palmas_pwron)
 	pm_wakeup_event(input_dev->dev.parent, 0);
 	input_sync(input_dev);
 
-	mod_delayed_work(system_wq, &pwron->input_work,
+	mod_delayed_work(system_dfl_wq, &pwron->input_work,
 			 msecs_to_jiffies(PALMAS_PWR_KEY_Q_TIME_MS));
 
 	return IRQ_HANDLED;
-- 
2.51.1

[PATCH v2 3/4] Input: mouse - replace use of system_wq with system_dfl_wq

[Marco Crivellari]

Currently if a user enqueues a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.

This lack of consistency cannot be addressed without refactoring the API.

This patch continues the effort to refactor worqueue APIs, which has begun
with the change introducing new workqueues and a new alloc_workqueue flag:

commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")

This specific workload do not benefit from a per-cpu workqueue, so use
the default unbound workqueue (system_dfl_wq) instead.

Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
---
 drivers/input/mouse/synaptics_i2c.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/input/mouse/synaptics_i2c.c b/drivers/input/mouse/synaptics_i2c.c
index a0d707e47d93..c8ddfff2605f 100644
--- a/drivers/input/mouse/synaptics_i2c.c
+++ b/drivers/input/mouse/synaptics_i2c.c
@@ -372,7 +372,7 @@ static irqreturn_t synaptics_i2c_irq(int irq, void *dev_id)
 {
 	struct synaptics_i2c *touch = dev_id;
 
-	mod_delayed_work(system_wq, &touch->dwork, 0);
+	mod_delayed_work(system_dfl_wq, &touch->dwork, 0);
 
 	return IRQ_HANDLED;
 }
@@ -448,7 +448,7 @@ static void synaptics_i2c_work_handler(struct work_struct *work)
 	 * We poll the device once in THREAD_IRQ_SLEEP_SECS and
 	 * if error is detected, we try to reset and reconfigure the touchpad.
 	 */
-	mod_delayed_work(system_wq, &touch->dwork, delay);
+	mod_delayed_work(system_dfl_wq, &touch->dwork, delay);
 }
 
 static int synaptics_i2c_open(struct input_dev *input)
@@ -461,7 +461,7 @@ static int synaptics_i2c_open(struct input_dev *input)
 		return ret;
 
 	if (polling_req)
-		mod_delayed_work(system_wq, &touch->dwork,
+		mod_delayed_work(system_dfl_wq, &touch->dwork,
 				msecs_to_jiffies(NO_DATA_SLEEP_MSECS));
 
 	return 0;
@@ -620,7 +620,7 @@ static int synaptics_i2c_resume(struct device *dev)
 	if (ret)
 		return ret;
 
-	mod_delayed_work(system_wq, &touch->dwork,
+	mod_delayed_work(system_dfl_wq, &touch->dwork,
 				msecs_to_jiffies(NO_DATA_SLEEP_MSECS));
 
 	return 0;
-- 
2.51.1

[PATCH v2 4/4] Input: psmouse-smbus - add WQ_UNBOUND to alloc_workqueue user

[Marco Crivellari]

Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.

alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.

This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.

This patch continues the effort to refactor worqueue APIs, which has begun
with the change introducing new workqueues and a new alloc_workqueue flag:

commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")

This specific workload do not benefit from a per-cpu workqueue, so use
WQ_UNBOUND on the custom workqueue instead.

With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.

Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.

Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
---
 drivers/input/mouse/psmouse-smbus.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/input/mouse/psmouse-smbus.c b/drivers/input/mouse/psmouse-smbus.c
index 93420f07b7d0..15bd49ccad22 100644
--- a/drivers/input/mouse/psmouse-smbus.c
+++ b/drivers/input/mouse/psmouse-smbus.c
@@ -299,7 +299,7 @@ int __init psmouse_smbus_module_init(void)
 {
 	int error;
 
-	psmouse_smbus_wq = alloc_workqueue("psmouse-smbus", 0, 0);
+	psmouse_smbus_wq = alloc_workqueue("psmouse-smbus", WQ_UNBOUND, 0);
 	if (!psmouse_smbus_wq)
 		return -ENOMEM;
 
-- 
2.51.1

Re: [PATCH v2 0/4] replace system_wq with system_dfl_wq, add WQ_UNBOUND to alloc_workqueue

[Dmitry Torokhov]

On Thu, Nov 06, 2025 at 03:19:51PM +0100, Marco Crivellari wrote:
> Hi,
> 
> === Current situation: problems ===
> 
> Let's consider a nohz_full system with isolated CPUs: wq_unbound_cpumask is
> set to the housekeeping CPUs, for !WQ_UNBOUND the local CPU is selected.
> 
> This leads to different scenarios if a work item is scheduled on an
> isolated CPU where "delay" value is 0 or greater then 0:
>         schedule_delayed_work(, 0);
> 
> This will be handled by __queue_work() that will queue the work item on the
> current local (isolated) CPU, while:
> 
>         schedule_delayed_work(, 1);
> 
> Will move the timer on an housekeeping CPU, and schedule the work there.
> 
> Currently if a user enqueue a work item using schedule_delayed_work() the
> used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
> WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
> schedule_work() that is using system_wq and queue_work(), that makes use
> again of WORK_CPU_UNBOUND.
> 
> This lack of consistency cannot be addressed without refactoring the API.
> 
> === Recent changes to the WQ API ===
> 
> The following, address the recent changes in the Workqueue API:
> 
> - commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
> - commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
> 
> The old workqueues will be removed in a future release cycle.
> 
> === Introduced Changes by this series ===
> 
> 1) [P 1-2-3]  Replace uses of system_wq
> 
>     system_wq is a per-CPU workqueue, these workloads do not benefit
> 	from locality, so change the the workqueue with the unbound,
> 	system_dfl_wq.
> 
> 2) [P 4] add WQ_UNBOUND to alloc_workqueue()
> 
> 	alloc_workqueue() users must specify one among WQ_PERCPU or WQ_UNBOUND.
> 	This workload does not benefit from a per-cpu workqueue, so make the
> 	custom wq unbound with WQ_UNBOUND.
> 
> Thanks!
> 
> ---
> Changes in v2:
> - all the patches use system_dfl_wq and the custom wq is now unbound,
>   using explicitly WQ_UNBOUND.
> 
> Marco Crivellari (4):
>   Input: gpio_keys - replace use of system_wq with system_dfl_wq
>   Input: palmas-pwrbutton - replace use of system_wq with system_dfl_wq
>   Input: mouse - replace use of system_wq with system_dfl_wq
>   Input: psmouse-smbus - add WQ_UNBOUND to alloc_workqueue user
> 
>  drivers/input/keyboard/gpio_keys.c    | 2 +-
>  drivers/input/misc/palmas-pwrbutton.c | 2 +-
>  drivers/input/mouse/psmouse-smbus.c   | 2 +-
>  drivers/input/mouse/synaptics_i2c.c   | 8 ++++----
>  4 files changed, 7 insertions(+), 7 deletions(-)
> 

Applied the lot, thank you.

-- 
Dmitry

Re: [PATCH v2 0/4] replace system_wq with system_dfl_wq, add WQ_UNBOUND to alloc_workqueue

[Marco Crivellari]

On Fri, Dec 19, 2025 at 5:54 AM Dmitry Torokhov
<dmitry.torokhov@gmail.com> wrote:
> Applied the lot, thank you.

Many thanks!



-- 

Marco Crivellari

L3 Support Engineer